TRWhisper

Name: TRWhisper — Download free. Local dictation
Availability: InStock
Rating: 5 (1 reviews)
Author: kagangungor

TRWhisper — Free Download. Local dictation

TRWhisper is a local push-to-talk dictation tool for Windows 11. Holding the Right Ctrl key records speech, and releasing it transcribes audio into text placed on the clipboard. The application operates without network connectivity, using the whisper.cpp engine, and supports optional NVIDIA GPU acceleration for faster processing. A floating overlay displays recording and transcription states, while a system tray icon provides access to model selection, transcript history, and settings.

★★★★★

5.0(1 ratings)

Download TRWhisper (Official links)

File size: 51.8 MB

The latest version of TRWhisper is: 2.2.0

Operating system: Windows

Languages: English, Turkish

Developer: kagangungor

Price: $0.00 USD (Open Source (MIT))

Global Push-to-Talk Hotkey. The application registers a low-level keyboard hook to detect the Right Ctrl key globally across the operating system. Holding the key starts audio capture from the default microphone, and releasing the key stops the capture and initiates transcription. The hotkey works regardless of which application currently has focus, enabling dictation into text editors, browsers, chat clients, and other software without switching windows manually.
Automatic Text Pasting. After transcription completes, the resulting text is placed on the Windows clipboard. The application then calls the Win32 SendInput API to simulate a Ctrl + V keystroke, inserting the text into the currently focused input field. The clipboard retains the text for additional manual pasting. This mechanism reduces the number of manual actions required to insert dictated text into a target application.
Offline Speech Recognition. All speech-to-text processing occurs on the local machine through a bundled whisper.cpp engine. Audio data is written to a temporary WAV file, processed by the engine, and then deleted immediately. No audio or transcribed text is transmitted to external servers during standard operation, which means the tool functions without an internet connection and keeps dictation content within the local environment.
NVIDIA GPU Acceleration. When the CUDA build of whisper.cpp is installed and an NVIDIA graphics card with a compatible driver is present, the application offloads model inference to the GPU. This reduces transcription time for short utterances from over twenty seconds on the CPU to roughly two or three seconds. If no usable GPU is detected, the application automatically falls back to CPU processing without interrupting functionality.
Whisper Model Switching. The system tray menu contains a submenu that lists available Whisper models, including Small and Large-v3 Turbo. Selecting a model updates the configuration file and takes effect on the next transcription without requiring an application restart. Models that are not present on disk appear greyed out. Choosing the Turbo model on a system without GPU support triggers a warning about potential slowness.
Silence Detection with Silero VAD. The integrated Silero voice activity detection model analyzes audio before transcription. Segments that contain no speech are excluded from processing. If the user presses the hotkey but does not speak, the application produces no output text instead of generating phantom phrases. This behavior reduces wasted computation and prevents unintended insertions.
Hallucination Filtering. Whisper models sometimes generate repetitive or nonsensical text when processing silence or noise. TRWhisper maintains a list of known hallucination patterns, such as subtitle credits and non-speech tags, and removes them from the transcript. The filter applies only when an entire line matches a known pattern, so legitimate dictation content remains unaffected.
Floating Pill Overlay. A small on-screen overlay indicates the current state of the application. During recording it displays a listening indicator. During transcription it shows a processing message. The overlay includes buttons to cancel the current operation or finish recording early. Once transcription completes, the overlay presents the resulting text along with a copy button for manual retrieval.
Optional LLM Cleanup. When the user holds Right Ctrl and Shift together, or enables the cleanup option from the tray menu, the raw transcript is sent to a configured Gemini or OpenAI API endpoint. The language model removes filler words and corrects punctuation before the text is pasted. This mode requires an API key and an internet connection, and it is disabled by default.
System Tray Controls. The notification area icon changes color to reflect application state: blue for idle, red for recording, and amber for transcribing. Right-clicking the icon opens a menu with options to toggle LLM cleanup, select a Whisper model, view the last ten transcripts, open the dictation folder, open the settings file, and exit the program.
Local Transcript Logging. Each completed transcription is appended to a monthly Markdown file stored in the user profile directory under a Dictation folder. Entries include a timestamp and the recognized text. Diagnostic messages are written to a separate log file. Temporary audio files are removed after processing to avoid accumulating recordings on disk.
Single-File Setup Wizard. The distributed installer bundles the application, the Silero VAD model, and a CPU Whisper engine. It offers optional downloads for the CUDA engine and additional models. The wizard detects NVIDIA hardware and recommends the GPU package when appropriate. Every download is verified with a SHA-256 checksum, and the installation requires no administrator privileges.

TRWhisper was created by kagangungor and first published on GitHub in September 2026. The project is written in C# using the .NET 9 framework and targets Windows 11 x64. Development began as a response to the lack of a local, privacy-oriented dictation tool tuned for the Turkish language. The repository contains setup scripts written in PowerShell, an Inno Setup installer definition, and source code organized under a src directory. The initial commit included clean source code and setup scripts, followed by additions for the installer, application icon, Turkish license page, and documentation updates with demo media and security verification information.